Developing predictive models for µ opioid receptor binding using machine learning and deep learning techniques

Abstract

Opioids exert their analgesic effect by binding to the µ opioid receptor (MOR), which initiates a downstream signaling pathway, eventually inhibiting pain transmission in the spinal cord. However, current opioids are addictive, often leading to overdose contributing to the opioid crisis in the United States. Therefore, understanding the structure-activity relationship between MOR and its ligands is essential for predicting MOR binding of chemicals, which could assist in the development of non-addictive or less-addictive opioid analgesics. This study aimed to develop machine learning and deep learning models for predicting MOR binding activity of chemicals. Chemicals with MOR binding activity data were first curated from public databases and the literature. Molecular descriptors of the curated chemicals were calculated using software Mold2. The chemicals were then split into training and external validation datasets. Random forest, k-nearest neighbors, support vector machine, multi-layer perceptron, and long short-term memory models were developed and evaluated using 5-fold cross-validations and external validations, resulting in Matthews correlation coefficients of 0.528-0.654 and 0.408, respectively. Furthermore, prediction confidence and applicability domain analyses highlighted their importance to the models' applicability. Our results suggest that the developed models could be useful for identifying MOR binders, potentially aiding in the development of non-addictive or less-addictive drugs targeting MOR.

Keywords: binding activity; deep learning; machine learning; predictive model; μ opioid receptor.

FIGURE 1

Study overview. The data on chemicals and their MOR binding activity were curated from public databases and the literature. The dataset from these databases was augmented with 1,727 non-binding chemicals sourced from the literature, forming the training dataset. The remaining chemicals from the literature constituted the external validation dataset. Molecular descriptors were calculated using Mold2 and subsequently filtered. Five algorithms—random forest, k-nearest neighbors, support vector machine, multi-layer perceptron, and long short-term memory—were used to build predictive models. The training dataset underwent 50 iterations of 5-fold cross-validation. Models constructed using the entire training dataset were then used to predict MOR binding on the external validation dataset. The performance of the models was evaluated based on their cross-validation and external validation predictions, with an additional focus on analyzing prediction confidence and applicability domain.

FIGURE 2

Performance of cross-validations. Performance of 50 iterations of 5-fold cross-validations was measured using sensitivity (A), specificity (B), balanced accuracy (C), accuracy (D), and MCC (E). The average values of these metrics across the 50 iterations are represented by color bars, corresponding to different algorithms indicated by the x-axis labels (RF, random forest; kNN, k-nearest neighbors; SVM, support vector machine; MLP, multi-layer perceptron; LSTM, long short-term memory; and CONS, consensus model). The standard deviations are displayed as error bars on top of the color bars.

FIGURE 3

Performance of external validations. The performance was assessed using sensitivity (A), specificity (B), balanced accuracy (C), accuracy (D), and MCC (E). The values of these metrics are represented by color bars for models developed using different algorithms, as indicated by the x-axis labels (RF, random forest; kNN, k-nearest neighbors; SVM, support vector machine; MLP, multi-layer perceptron; LSTM, long short-term memory, and CONS, consensus model).

FIGURE 4

Prediction confidence analysis results. The analysis of prediction confidence is depicted by plotting prediction accuracy values and the number of predictions at various confidence levels. (A, B) show the results for cross-validations, while (C, D) display the results for external validations. The x-axis tick labels represent the different confidence levels. The models developed using different algorithms are distinguished by various colors, as indicated in the color legend (RF, random forest; kNN, k-nearest neighbors; SVM, support vector machine; MLP, multi-layer perceptron; LSTM, long short-term memory, and CONS, consensus model).

FIGURE 5

Applicability domain (AD) analysis results. The AD analysis is presented for cross-validations (A) and external validations (B). Accuracy for predictions within the AD is shown in cyan bars, while accuracy for predictions outside the AD is displayed in orange bars. The models developed using different algorithms are represented by the x-axis labels (RF, random forest; kNN, k-nearest neighbors; SVM, support vector machine; MLP, multi-layer perceptron; LSTM, long short-term memory, and CONS, consensus model).